PGTS: New hybrid framework beats multi-agent game benchmarks on real graphs

Computing Nash equilibrium policies in multi-agent pursuit-evasion games (PEG) is notoriously difficult due to the exponential growth of joint state and action spaces as the number of agents increases. Existing methods fall into two camps: offline equilibrium approximations that lack adaptability during execution, and online planning approaches that suffer from enormous branching factors. In a new paper from arXiv, researchers propose Primitive-Guided Tree Search (PGTS), a hybrid framework that bridges this gap. PGTS first solves a collection of smaller, tractable sub-games offline using exact Nash equilibrium computation. Then, during deployment, PGTS performs online tree search at each time step, using the optimal sub-game policies and value functions to guide tree expansion and estimate leaf-node values.

Extensive experiments across diverse graph topologies—including real-world networks—demonstrate that PGTS significantly outperforms state-of-the-art learning and heuristic baselines. The framework maintains robust performance even when faced with adversarial opponents, making it a promising tool for applications in robotics, security, and autonomous multi-agent coordination. By combining the precision of offline equilibrium computation with the adaptability of online planning, PGTS offers a practical solution to the scalability challenges inherent in multi-agent games on graphs.